studies with different followup periods e.g a total of 5 RCTs (2 of them followed at 1 year) , (2 at three years ) and (one vat at 5 years) what is the best method to pool the data of these studies?
the measure of effect size include mean difference for continuous outcome
and risk for dichotomous outcomes.
Whether this is sensible depends on the assumptions you make about the data. Combining studies with different follow-up lengths is commonly done in meta-analysis, but rarely thought about properly. If the hazards are proportional (i.e. the difference in effect does not vary between the two groups over time) then you can use the methods described by Combescure et al:
Combescure C, Courvoisier DS, Haller G, et al. Meta-analysis of binary outcomes from two-by-two tables when the length of follow-up varies and hazards are proportional. Stat Methods Med Res. 2011;20:531–540.
In order to test your assumptions, as Rohit has suggested, you should plot effect size against follow-up time for all of your studies and look to see if there is a link between the two. Presence of heterogeneity should also alert you to a possible problem combining studies.
I agree with Rohit P Ojha. Exploring the potential influence of the follow-up duration on estimates is critical. Cooper (1989, pp. 88-90) discusses and illustrates two contrasting studies in which the time-treatment interaction causes treatment effects that (a) dissipate over time and (b) reverse themselves over time. In other words, follow-up duration needs to be statistically controlled whenever comparing studies with varying follow-up duration. Admittedly, this is a difficult task.
1. Cooper, H.M. ((1989), pp. 88-90). Integrating research: a guide for literature reviews (2nd ed.). Newbury Park, London, New Delhi: Sage Publications.
Dear Ahmed, in our recently published paper [1] we tackle the same question. We work with competing risks but the idea can be applied to a single outcome type. I assume you are working with aggregated survival data and you wish to measure the treatment effect on the hazard scale. If this is the case, we argue it is wise to also account for follow-up lengths differences between studies.
In order to account for time differences between studies we use a sort of jack-knife approach. On a continuous time line, t=(0, tau), where tau is typically the maximum observed follow-up among all studies, a study is pooled only if its follow-up duration is at most equal to time t. That is, at each study specific follow-up time a study is dropped out of the pooling process, because it does not contribute observed evidence beyond that time.
Such approach is best appreciated graphically, by plotting the value of the pooled HR against time. If the pooled HR wobbles a great deal at each dropped study, this may indicate that the assumption of constant (or proportional) hazards is not tenable across studies. Alternatively, it may also means that there is some time-related confounding from one study to another and this is may affect heterogeneity. I hope I was clear enough, but I suggest you give a look at our paper [1] because we had not found much literature on your question at that time either.
[1] Bonofiglio et al. (2015) Meta-analysis for aggregated survival data with competing risks: a parametric approach using cumulative incidence functions. Journal of Research Synthesis Methods, Wiley.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis